A Vector Autoregressive (VAR) Model for the Hourly Forecasting of Climatological Data in General Santos City

Proponent/s Name/s (Last Name, First Name, Middle Initial)

Jezzu Morrisen C. Quimosing, Mindanao State University-General Santos City, Senior High School DepartmentFollow
Feevol S. Into, Mindanao State University-General Santos Senior High SchoolFollow
Ayeesha Kate N. Castro, Mindanao State University-General Santos Senior High SchoolFollow
Criss Allyn C. Hongoy, Mindanao State University-General Santos Senior High SchoolFollow
Bai Ashfahana A. Abdullah, Mindanao State University-General Santos Senior High SchoolFollow
Ma. Sofia Franchesca P. Dizor, Mindanao State University-General Santos Senior High SchoolFollow
Leo James M. Podador, Mindanao State University-General Santos Senior High SchoolFollow
Joshua Caleb G. Pagalan, Mindanao State University-General Santos Senior High SchoolFollow
Ralph Laurence Valdueza, Mindanao State University-General Santos Senior High SchoolFollow

Document Type

Paper Presentation

Research Advisor (Last Name, First Name, Middle Initial)

Valdueza, Ralph Laurence

Abstract/Executive Summary

With rising concerns over climate change and its impacts, this study develops and evaluates a Vector Autoregressive (VAR) model to forecast climatological data in General Santos City: surface pressure, specific humidity, wind speed, air temperature, precipitation, UVA and UVB irradiance, with data from January 1, 2001 to June 30, 2024. Mean hour calculations confirmed seasonality via the QS test (p < 0.001) for all variables except precipitation; Mann-Kendall tests indicated significant trends for all tested variables, and UV irradiances were the only variables with no significant long‐term trend. The Augmented Dickey-Fuller Test (p < 0.001) established stationarity of variables, and Granger‐causal relationships were identified. Results indicated a bidirectional Granger-causal relationship between variable pairs (p < 0.001). The AIC-based model selection identified VAR(168) as the best-fitting model, effectively capturing cyclic patterns up to one week. The residuals passed the Ljung-Box tests for no residual autocorrelation (p > 0.05), and OLS-CUSUM plots indicated structural stability. Mean root-mean-square error determined out-of-sample forecasting accuracy to be good for most variables, and poor for UVA and UVB irradiance; therefore, while the VAR(168) model does generate accurate forecasts, it still presents challenges. With this, local organizations can make timely decisions related to agricultural adjustments and precautions, disaster preparedness activities, and public health initiatives. It is suggested that future studies should use ensemble forecasting techniques, test for nonlinear VAR extensions, and compare VAR results with physical-statistical models to enhance precipitation forecasts in tropical regions further.

Keywords

climate forecasting; General Santos City; multivariate time series analysis; NASA Power API; vector autoregressive (VAR) model